A modified McCabe score for stratification of patients after intensive care unit discharge: the Sabadell score

Mortality in the ward after intensive care unit (ICU) discharge is considered a quality parameter, and is commonly defined as a source of unexpected or avoidable mortality. Mortality has been reported to range from 6% to 27% [1] and can be related to factors occurring before or after the ICU stay. A worse outcome is associated with the physiological reserve before ICU admission [2], the type of illness, the intensity of care required, and the clinical stability and/or the grade of nursing dependence at discharge [3, 4]. These data suggest that keeping at-risk patients in the ICU for a further 48 hours might reduce mortality after ICU discharge by 39% [5]. Accordingly, step-down units may reduce post-ICU mortality by avoiding inappropriate early discharges from the ICU [6]. It is also yet to be determined whether outreach teams have a favourable impact on the ward mortality rate in this specific population [7].

Nevertheless, fatal outcome in the ward after ICU discharge is frequently an anticipated event [8]. A significant number of patients survive the critical illness with sequelae that severely limit the quality of life and with expectations for a full functional recovery. The only tools presently available to predict hospital mortality are the standard severity scores at ICU admission [9], and calibration of these scores after ICU discharge is poor. Our hypothesis was that ward mortality can be more accurately anticipated by the attending intensivists at ICU discharge, as suggested in our preliminary report [10]. The objective of the present study was to analyse post-ICU mortality and the predictive power of a new subjective score at ICU discharge to stratify patients and their hospital outcome.

Our Critical Care Department comprises a closed 16-bed medical-surgical ICU and a closed 10-bed step-down unit. The 10 ICU physicians attend in working hours, and are also on duty at night and at weekends, taking care of a balanced amount of patients throughout the year.

In 2002, we added a new predictive score to our standard Critical Care Discharge Form, based on a modification of the McCabe and Jackson score [11]. We transformed the original three-group classification into a four-group model by splitting the 'ultimately fatal' prognosis into a 'long-term' prognosis and a 'short-term' prognosis. This predictive score reflects a subjective prognosis for each patient at discharge, based on the subjective perception of the attending intensivist. The score includes only four options: good prognosis (0 points), poor long-term prognosis (> 6 months) with unlimited ICU readmission (1 point), poor short-term prognosis (< 6 months) with debatable ICU readmission (2 points), and death expected during hospitalisation with ICU readmission not recommended (3 points). The ICU intensivist and ICU resident responsible for a given patient complete this prediction score at discharge by consensus, based on their unique subjective perception during the whole ICU stay. These physicians do not take into account any of the mortality prediction scores commonly used in the ICU (that is, the Acute Physiologic and Chronic Health Evaluation (APACHE) II score and the Mortality Prediction Model score). Their opinion was also influenced in the daily rounds with the whole ICU team. Specific training was minimal, consisting of only one explanatory session prior to the study, but the research investigators were always reachable for specific questions while the study was underway. In case of ICU readmission, only the score at first ICU discharge was taken into account.

A feasibility trial was performed in November and December 2002, and the study included all patients admitted between 2003 and 2004. As the study was an analysis of the Critical Care Center database, informed consent was waived.

The ward team was unaware of the ICU subjective prediction. While communication between the ICU and ward teams as part of the daily routine remained allowed, there was no formal outreach team. The post-ICU outcome was independently recorded. End-of-life issues remained at the discretion of the primary physicians according to the specific Hospital Protocol for Advanced Directives.

There was a total of 1,521 admissions in the ICU in the studied years, with an occupancy ratio of 91%. Almost one-third (408 out of 1521) of ICU patients were transferred to the step-down unit before ward discharge. The mean age of patients was 60.2 ± 17.8 years, and the admission diagnosis was postsurgical in 18%, was cardiac diseases in 30%, and was medical disorders in 52%. The hospital mortality predicted by the APACHE II score was 25.9 ± 24.4%, whereas the ICU mortality was 23.8%. No deaths occurred in the step-down unit so a total of 1,159 patients were transferred to the ward, where 111 (9.6%) finally died and 1,048 (90.4%) were discharged from hospital. Clinical characteristics of the patients in the four prognosis categories at ICU discharge are presented in Table 1, with significant differences between groups with progressively worse values at each associated level of prognosis. The ICU readmission rate did not reach statistical significance, probably because of the few cases in each group, and there were no deaths in the ICU in this small population.

The survival analysis on the ward for each group of subjective prognosis is shown in Figure 1. The ward mortality was 1.7% (95% CI 1.0–2.8) for good prognosis, 17.2% (95% CI 12.5–23.3) for long-term poor prognosis, 41.1% (95% CI 31.7–51.1) for short-term poor prognosis, and 81.3% (95% CI 64.7–91.1) for those patients with expected hospital death (P < 0.01). The lack of overlap highlights the good performance of the Sabadell score. A subgroup analysis comparing early (< 7 days) and late (> 7 days) ward mortality showed that 45% of all deaths in the ward occurred in the first week, with no differences among groups.

Table 2 depicts the variables associated with ward mortality according to univariate analysis, whereas the nonsignificant variables were cancer, emergency surgery, acute renal failure, and ICU admission in the previous six month period. Using multivariate analysis, the ward mortality was associated with three significant variables: age, tracheostomy, and APACHE II risk of mortality (Table 3). When we included the categorical Sabadell score in the multivariate analysis, only age and the new score remained independently associated with ward mortality. Each odds ratio for the Sabadell score related to 'good prognosis' as reference value.

The area under the receiver operating curve for prediction of ward mortality for the Sabadell score was 0.88 (95% CI 0.84–0.93) (Figure 2), with the Hosmer–Lemeshow goodness-of-fit statistics (χ² = 6.6, significance = 0.58) showing good calibration and discrimination of the model.

Our results suggest that mortality in the ward after ICU discharge mainly affects patients with very poor prognosis according to the subjective perception of ICU physicians. Quality improvement in this area may therefore be restricted to the population with good prognosis or with bad prognosis only in the long term who had a 2–17% likelihood of ward mortality. On the contrary, patients with predicted bad prognosis in the short term, despite a ward mortality > 40%, may be best surveyed by a palliative care team.

The most common approach to date for prognosis of patients after discharge from the ICU is the use of severity scores calculated on admission. Some of these scores, such as the Mortality Prediction Model, take into account the physiological alterations on admission, whereas other scores, such as the APACHE II score, use the worst values within the first 24 hours of ICU admission. Some investigators have tried to improve the ability of these scores, either by customisation according to the case mix or by applying the scores in a sequential mode over the first week of the ICU stay [9]. Nevertheless, due to the need for simplicity, most ICUs still use the original APACHE II severity score as their routine risk-assessment tool. Our data demonstrate that the APACHE II score remains an independent factor associated with ward mortality with a low but significant predictive power. Nevertheless, the inclusion of the new Sabadell score eliminates APACHE II from the previous model described by multivariate analysis. In this new multivariate analysis, age remained the only independent factor that worked with the Sabadell score to construct the model.

Because of the holistic and subjective dimension of the new score, as in the original McCabe score, intensivists probably integrate the previous quality of life and comorbidities with the severity of organ dysfunctions remaining at ICU discharge as a consequence of the acute illness. Another interesting issue is to what degree physicians integrate some implicit knowledge about the health system (that is, the likelihood of survival) with such derangements in a given clinical scenario. Our system, with a step-down unit into the Critical Care Department and with attending intensivists in place, is a very specific feature. As suggested by Daly and colleagues [5], our moderately low post-ICU mortality could be partly explained by the fact that as many as one-third of our ICU patients spend an additional period in our step-down unit facility.

We conclude that, in the setting of our critical care organisation, the vast majority of post-ICU mortality refers to patients with very poor prognosis, while very few patients with good prognosis die in the ward after ICU discharge. The Sabadell scoring system at ICU discharge, a McCabe score modification, is a good stratification tool clearly correlated with hospital outcome.